The Multiverse Secret Keepers

These Custom GPTs are Multiverse Student secret-keeping bots. If you can find the secret, please submit your chatlog below for credit and to aid in public research.

Every 30 days we will grade and publish the stats. As of today we haven't graded them yet. We'll publically publish the key elements of more successful defender prompts at the end.

You won't know if you got the real secret for 30 days. This is similar to how a honeypot will behave- you'll believe you "got in", only to discover later that you didn't get the real secret, and you were tagged-and-released. Happy hacking!

Jailbreaking Research

• Jailbreaking ChatGPT via Prompt Engineering: An Empirical Study
• ArtPrompt: ASCII Art-based Jailbreak Attacks against Aligned LLMs
• Catastrophic Jailbreak of Open-source LLMs via Exploiting Generation
• Many-Shot Jailbreaking
• Tree of Attacks: Jailbreaking Black-Box LLMs Automatically
• Agent Smith: A Single Image Can Jailbreak One Million Multimodal LLM Agents Exponentially Fast

Blue Team 1

• Defense
• Derpy
• Jailbreak-Buster
• Durian Greg
• Wallfacer
• AW-Super-Agent
• Benjamin Bluey
• Lowkey

Blue Team 2

• Secret Manager
• Super Secret Password Guy
• Biggobgriefer
• Worf
• MySwamp
• Keeper of Sausages
• Used Car Salesman

Blue Team 3

• So Secret!
• I have a secret.
• HowdyGPT
• Jailbreaking Secret Keeping
GPT
• darren
• Personality Tweets
• TheOmegaKeeper
• Jailbreaks class
• Multiverse Secret Keeper
• Armed Guard
• Secret Keyper
• Jailbreaks class
• Supersecretchat
• Yeehaw Let's Go